The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes git diff to identify and gather code changes for review. This is a standard developer workflow operation and relies on local repository data.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code and diff data (SKILL.md, Step 1). This ingestion surface is mitigated by the use of a structured prompt template (reviewer-prompt.md) for subagents and a pragmatic lead-judgment synthesis framework (lead-judgment.md) that filters and weights findings. No explicit sanitization of code content is performed before interpolation into subagent prompts.
[DYNAMIC_EXECUTION]: Reviewer subagents are dynamically instantiated using the Task tool. The configuration for these subagents (models and reasoning tiers) is retrieved from a local configuration file at ~/.cursor/rules/pstack-models.mdc or defaults specified in the skill instructions. Subagents are specifically restricted to readonly: true to prevent unauthorized modifications to the environment during the review process.

interrogate